What buyers ask before they buy a video agent
The inbound email has changed shape. It now arrives as a numbered list: false-positive governance, human-in-the-loop, auditability, on-prem options, validation paperwork, and a pilot whose length the buyer has already decided.
A year ago the typical inbound request for a video agent was a sentence long. Someone had seen a demo, wanted one for their own site, and asked for a call. The requests I have been reading from the past few months look different. They arrive as numbered lists, and the numbers run past ten.
The list
The clearest example came from the VP of business strategy at a global animal-health manufacturer. His email opened with the scope, "real-time exception detection, worker safety, and operational oversight across our facilities," and then set out ten items he wanted answered before a demo. Several would have been unusual in a security RFP two years ago.
Item two was human-in-the-loop, defined as "How low-confidence or high-risk exceptions are routed to human reviewers." Item three was auditability: "Whether alerts can be traced to source frame, timestamp, model version, reviewer, and disposition." Item four asked for "On-premise or edge options, with no cloud upload of production video unless approved." Item six was false-positive governance, "How errors are measured, tuned, and managed over time." Item eight asked for validation support, meaning "Documentation available for regulated or GMP-adjacent environments."
Read those together and you have a buyer who assumes the model will be wrong some of the time and wants to know what the vendor does about it. Model version and reviewer disposition on every alert is what a quality team asks for. In a GMP-adjacent plant, an alert that changes how a line runs is a record, and records need a chain of custody.
Smaller buyers send the same list
A franchisee running six quick-service dessert stores sent nine numbered capabilities and asked the vendor to mark each one "supported today, supported with configuration, on roadmap, or not supported" before any demo was scheduled. He wanted ballpark pricing for "a typical 4-8 camera store," said he had existing IP cameras at most sites and one point-of-sale system at all of them, and closed with a line for the sales team: "I'm evaluating one other vendor in parallel."
An investor sitting through a diligence demo asked the question underneath all of these: whether the AI has to be retrained for every machine variation and every location, and how often a customer has to call the vendor when a standard operating procedure changes, or whether they can fix it themselves.
The pilot is already designed
The other change is that buyers now arrive with the pilot shape decided. Three shapes come up.
The first is eight weeks at a single station. A laboratory equipment manufacturer had run a pre-sale analysis with the vendor that, in the buyer's words, "showed that roughly 90% of our reworks occur at just a few stations." The executive team's response was to shrink the scope: "the team would like to run a simple, turnkey proof-of-value at a single station for 8 weeks, with Spot AI providing the cameras." The evaluation criteria were in the same email: "Functionality, Detection rates, Reporting, Data management." He asked for "a no-commitment quote for this single-station trial" and for a recommendation on which station to pick. The larger order, with an option to expand to 20 cameras the following year, had been reviewed and set aside until the trial ran.
The group remains excited about the direction — we just want to see it perform firsthand before determining how to scale.
The second shape is 30 days at one or two stores, which is what the dessert franchisee proposed "before a broader rollout." The third is the animal-health manufacturer's tenth item: "What a structured 90-day pilot would look like for one use case at one of our sites."
Vendors and partners are less happy about this. A channel partner working a thousand-location opportunity was recorded saying "Nobody's making money with them having a handful of cameras in a freaking pilot," and hoped to get the customer to commit while the pilot was still running. A national integrator group wrote in its own recap of a food-and-beverage manufacturer's project that "Charging for the build phase instead of proving it out free landed better." An operator of licensed coffee stores told the vendor that any proof of concept would need defined success metrics and sample good and bad clips for training up front, and might have to be a paid engagement.
So the two sides are converging on a paid trial with a definite shape: one station or one store, a fixed number of weeks, criteria written down in advance, and a price attached. The lab equipment manufacturer's version gives the vendor eight weeks and one station, and the larger order waits on the result.